strata generate: PCIe probe: 29.4 GB/s host->device (best of 25.6 25.6 26.0 29.4) -> pcie_frac 0.55 (default 0.55)
strata generate: native pack: /data/packs/qwen-q2_0 experts (largest blob 1.38 MB), token embedding Q3_K in mapped host memory (260 MiB, 0.2 s)
strata generate: 1416 MiB of weights loaded from /data/packs/qwen-q2_0 in 0.6 s (303 canonical tensors skipped: served natively)
strata generate: 301 native projection matrices, 1376.20 MiB of weights, in 0.7 s
strata generate: PLE direct read probe: 67528 rows/s (fine)
strata generate: the SSD is kept awake while rows are read: one page of the table after 100 ms without a read, until 60 s after the last request (STRATA_SSD_KEEPALIVE=0 turns it off)
strata generate: profile /src/data/expert-profile.bin: 24576 ranked pairs, built for 24576 slots
strata generate: PLE on, table 320001536 rows of /models/flash-next/Q2_0/Qwen3.8-Flash-Next-GSQ-RCO-Q2_0-00002-of-00002.gguf
strata mtp: draft layer loaded, 799 MiB of VRAM (experts 675, dense 111), files read in 0.47 s (1654 MiB/s)
strata generate: experimental native Q5_K head, 437043200 bytes, in 0.2 s
strata generate: GPU 0: AMD Radeon 780M Graphics (gfx1103)
strata generate: experts via mmap (--mmap-experts; the GGUF shards in place, no experts.bin)
strata generate: GPU host-callback thread started before host-core pinning
strata generate: CPU pool tasks/phase: 24 (automatic), participating threads: 8
strata generate: 7 pool workers on logical processors 1,2,3,4,5,6,7, the host thread on 0 (draining too) (--host-core first)
strata generate: expert cache 24576 slots, 31.64 GiB of VRAM; policy is
strata generate: the GPU computes the experts in the cache; it rounds differently from the CPU,
strata generate: the profile fill asks for 256 pairs ahead (STRATA_FILL_AHEAD)
strata generate: pre-filled 24576 of 24576 slots from the profile in 21.9 s (1550 MB/s); slot 0 verified
strata generate: routing-aware prefetch on (the next layer's router, file tier)
strata generate: R4 hit path ON - resident experts are computed on the GPU
strata generate: 7 expert-pool workers + the host thread
strata generate: session is up (engine 0.1.43.1)
strata generate: sampling greedy
strata generate: token graph hit path: 24576 resident experts, decided on the device
strata generate: prompt chunk auto: 256 tokens
strata generate: prompt path borrows 191 cache slots (0.25 GiB)
prefill gemm: hipBLASLt tuning enabled (143 rows, gfx1103, version 100401)
strata generate: 191 lent slots refilled in 23.4 ms
strata generate: prefill 1 tokens in 1 chunks, 729.8 ms (1.4 tok/s); experts streamed 0 (0 by DMA, host 0.0 ms), resident 480, on the CPU 0 (share 0.00); PLE 0.9 ms
strata verify: window up to 6 tokens, 54.8 MiB of device buffers (100% VRAM resident: zero-doorbell graph)
strata verify: capturing the 1-token window (14137 MiB of VRAM free)
strata verify: captured the 1-token window (upload no error, sync no error)
strata verify: capturing the 2-token window (14134 MiB of VRAM free)
strata verify: captured the 2-token window (upload no error, sync no error)
